Ultrahigh resolution MS1/MS2-based reconstruction of metabolic networks in mammalian cells reveals changes for selenite and arsenite action

Metabolic networks are complex, intersecting, and composed of numerous enzyme-catalyzed biochemical reactions that transfer various molecular moieties among metabolites. Thus, robust reconstruction of metabolic networks requires metabolite moieties to be tracked, which cannot be readily achieved with mass spectrometry (MS) alone. We previously developed an Ion Chromatography-ultrahigh resolution-MS1/data independent-MS2 method to track the simultaneous incorporation of the heavy isotopes 13C and 15N into the moieties of purine/pyrimidine nucleotides in mammalian cells. Ultrahigh resolution-MS1 resolves and counts multiple tracer atoms in intact metabolites, while data independent-tandem MS (MS2) determines isotopic enrichment in their moieties without concern for the numerous mass isotopologue source ions to be fragmented. Together, they enabled rigorous MS-based reconstruction of metabolic networks at specific enzyme levels. We have expanded this approach to trace the labeled atom fate of [13C6]-glucose in 3D A549 spheroids in response to the anticancer agent selenite and that of [13C5,15N2]-glutamine in 2D BEAS-2B cells in response to arsenite transformation. We deduced altered activities of specific enzymes in the Krebs cycle, pentose phosphate pathway, gluconeogenesis, and UDP-GlcNAc synthesis pathways elicited by the stressors. These metabolic details help elucidate the resistance mechanism of 3D versus 2D A549 cultures to selenite and metabolic reprogramming that can mediate the transformation of BEAS-2B cells by arsenite.

Metabolic networks are complex, intersecting, and composed of numerous enzyme-catalyzed biochemical reactions that transfer various molecular moieties among metabolites. Thus, robust reconstruction of metabolic networks requires metabolite moieties to be tracked, which cannot be readily achieved with mass spectrometry (MS) alone. We previously developed an Ion Chromatography-ultrahigh resolution-MS 1 /data independent-MS 2 method to track the simultaneous incorporation of the heavy isotopes 13 C and 15 N into the moieties of purine/pyrimidine nucleotides in mammalian cells. Ultrahigh resolution-MS 1 resolves and counts multiple tracer atoms in intact metabolites, while data independent-tandem MS (MS 2 ) determines isotopic enrichment in their moieties without concern for the numerous mass isotopologue source ions to be fragmented. Together, they enabled rigorous MS-based reconstruction of metabolic networks at specific enzyme levels. We have expanded this approach to trace the labeled atom fate of [ 13 C 6 ]-glucose in 3D A549 spheroids in response to the anticancer agent selenite and that of [ 13 C 5 , 15 N 2 ]-glutamine in 2D BEAS-2B cells in response to arsenite transformation. We deduced altered activities of specific enzymes in the Krebs cycle, pentose phosphate pathway, gluconeogenesis, and UDP-GlcNAc synthesis pathways elicited by the stressors. These metabolic details help elucidate the resistance mechanism of 3D versus 2D A549 cultures to selenite and metabolic reprogramming that can mediate the transformation of BEAS-2B cells by arsenite.
Metabolomics has been instrumental in accelerating the elucidation of metabolic reprogramming induced by disease states or drug treatment (1)(2)(3)(4) and the discovery of metabolism-based biomarkers (5)(6)(7). As metabolite levels are governed by many factors including rates of synthesis and degradation, multiple input and output pathways, and exchange across compartments (8), it has been challenging to reconstruct metabolic networks based on total metabolite profiles alone. Metabolic transformations through the network require numerous enzyme-catalyzed reactions that transfer the structural moiety among metabolites. Thus, the ability to track metabolite moiety will greatly reduce the ambiguities in metabolic network analysis. Stable isotope-resolved metabolomics (SIRM) fulfills this requirement by systematically tracking the transformations of individual tracer atoms from precursors to products using a combination of MS 1 and NMR methods, which provides respectively the number and position of the tracer atoms in given metabolites. This approach has been successfully applied to determine altered metabolic activities by disease development and other perturbations in 2D/3D cell cultures (9)(10)(11)(12)(13)(14)(15), human tissues ex vivo (2,16,17), patient-derived xenograft mice in vivo (18,19), and even human subjects in vivo (2,20).
However, compared with mass spectrometry (MS), the moderate sensitivity of NMR limits the overall metabolite coverage. This limitation prompted us to develop an Ion Chromatography-Ultrahigh Resolution-MS 1 /data independent-MS 2 (IC-UHR-MS 1 /DI-MS 2 ) method to enable determination of tracer atom position(s) in metabolite moiety by MS with higher resolution and sensitivity than NMR. This in turn allows robust reconstruction of metabolic network responses to stressors at specific enzyme levels (21). The UHR-MS 1 step is capable of resolving the neutron mass difference among different tracer atoms (e.g., Δmass = 0.006995 amu between 13 C and 15 N) (10,22). This capability enables multiplexing of biologically compatible tracer atoms such as 13 C, 15 N, and 2 H in the same (e.g., [ 13 C 5 , 15 N 2 ]-Gln) or different substrates (e.g., [ 13 C 6 ]-glucose + [ 15 N 2 ]-Gln) to expand the metabolic pathway coverage while circumventing sample batch effects in multiplex SIRM studies (10,23).
We have expanded the pathway reconstruction of purine/ pyrimidine nucleotide synthesis to the reconstruction of metabolic networks consisting of the Krebs cycle, pentose phosphate pathway (PPP), gluconeogenesis, and UDP-GlcNAc synthesis pathways in 3D A549 spheroids and arsenitetransformed BEAS-2B cells. By tracing [ 13 C 6 ]-glucose or [ 13 C 5 , 15 N 2 ]-Gln transformations into the moiety of these pathway metabolites, we were able to deduce changes in specific enzyme activities induced by selenite in A549 spheroids or by arsenite in BEAS-2B cells. This information enabled us to surmise the resistance mechanism of 3D versus 2D A549 cultures to selenite and metabolic reprogramming that presumably mediates the transformation of BEAS-2B cells by arsenite.

Results
Isotope enrichment distributions of major metabolites from glycolysis, the Krebs cycle, PPP, gluconeogenesis, and UDP-GlcNAc metabolism were obtained from the UHR-MS 1 and MS 2 spectra in both [ 13 C 6 ]-glucose-traced A549 spheroids ± anticancer selenite treatment and [ 13 C 5 , 15 N 2 ]-Gln-traced BEAS-2B cells compared with arsenite transformated BEAS-2B cells. Example MS 1 (A) and MS 2 (B) spectra are shown for citrate in Fig. S1. Isotopologue concentrations were calculated from the peak area ratio of samples to calibration standard mixtures after natural abundance correction, followed by normalization to the sample protein concentration.

The Krebs cycle
The glycolytic product of [ 13 C 6 ]-Glc ( 13 C 3 -pyruvate) enters the Krebs cycle either via 13 C 2 -acetyl CoA produced from the pyruvate dehydrogenase (PDH) reaction or directly into 13 C 3oxaloacetate via pyruvate carboxylase (PC) activity. After the first turn, the PDH-initiated Krebs cycle produces 13 C 2 -isotopologues ( ) of various intermediates, whereas PCB-initiated Krebs cycle generates 13 C 3 -isotopologues ( ) of citrate, cisaconitate, malate, fumarate, and aspartate (2), and the malic enzyme (ME) reaction scrambles 13 C in pyruvate leading to the synthesis of 13 C 1 -metabolites ( ) (Figs. S2A and 1A). It should be noted that this pathway scheme takes into account unlabeled carbon ( • ) that can come from preexisting pools of free metabolites as well as their precursors such as glycogen, proteins, and lipids.
In the [ 13 C 6 ]-Glc-traced A549 spheroids, we saw the occurrence of 13 C 2 -(2, red box) and 13 C 3 -citrate (3, green box), which are the respective products of PDH-initiated (canonical) and anaplerotic PC-initiated Krebs cycle (Figs. 1A-b and S1A). The presence of the 13 C 3 -3,4,5-citrate species (3) in the MS 2 data also points to PC activity (Figs. 1A-c and S1B). It is evident from the UHR-MS 1 data that 13 C 2 -citrate accumulated more than 13 C 3 -citrate, indicating a higher activity of PDH-initiated than anaplerotic PC-initiated Krebs cycle. However, 13 C 2 -malate (f) and -Asp (g) were comparable in levels to the 13 C 3 -counterparts (Fig. 1A). This discrepancy can be accounted for by the contribution of a second turn canonical Krebs cycle activity to the 13 C 3 pools, which is consistent with the synthesis of 13 C 4 -citrate (b), a specific product of the second turn. Although low in levels, 13 C 1 -citrate and -Glu (i) were present, suggesting contribution from the ME reaction. Selenite induced the depletion of all 13 C 2 -, 13 C 3 -, and 13 C 1 -isotopologues of the Krebs cycle intermediates in A549 spheroids, except for αKG (d), which showed enhanced buildup. These data are consistent respectively with inhibition of PDH, PC, and ME-mediated Krebs cycle activity, particularly at the α-ketoglutarate dehydrogenase (OGDH) step by selenite leading to the accumulation of all 13 C-isotopologues of αKG. The 13 C-labeling patterns of the MS 2 fragments verified the selenite effect on PDH (2 or 13 C 2 -1,2-Asp, h; 3 or 13 C 3 -Glu-GSH, k) and PC (3 or 13 C 3 -1,2,3-Asp, h) activity (Fig. 1A) while revealing inhibition of GSH synthesis by blocking the PDH-initiated Krebs cycle activity and Ser→Gly synthesis pathways (cf., Fig. S3). The latter is evidenced by the depletion of 13 C 3 -Glu (k) and 13 C 2 -Gly (l) moiety of GSH. This information could not be ascertained based on the MS 1 data of GSH (j) alone (Fig. 1A).
In [ 13 C 5 , 15 N 2 ]-Gln-traced BEAS-2B cells, the labeled Gln enters the Krebs cycle by first conversion to 13 C 5 , 15 N 1 -Glu (a) via glutaminase-catalyzed glutaminolysis and then to 13 C 5 -αKG (b) via glutamic-oxaloacetic transaminase (GOT)-catalyzed transamination and/or glutamate dehydrogenase 1-catalyzed oxidative deamination. 13 C 5 -αKG is further transformed to 13 C 4 -succinate (d), -fumarate (e), -malate (f), and -citrate (h) via the Krebs cycle (Figs. S2B and 1B). 13 C 4 -malate can be converted to 13 C 3 -pyruvate (l) via the ME reaction, leading to the synthesis of 13 C 2 -and 13 C 3 -citrate, -succinate, -fumarate, -malate, and -Asp via, respectively, PDH-and PCinitiated Krebs cycle activities. Moreover, 13 C x , 15 N-Asp (j) can be produced via GOT-catalyzed transamination while 13 C x , 15 N-GSH (m) is synthesized from 13 C x , 15 N-Glu. Such pathway reconstruction was deduced from the presence of all expected 13 C and 13 C, 15 N-isotopologues of the glutaminolytic and Krebs cycle products based on the MS 1 and MS 2 data. Arsenite transformed cells (BAsT) showed depletion of all of these products except for the labeled GSH in terms of both Glu (n) and Gly (o) moieties (Fig. 1B). These data pointed to inhibition of the glutaminase and/or Krebs cycle activity but activation of GSH synthesis in BAsT versus control cells.
In addition, detailed analysis of the 13 C-and/or 15 N-labeling patterns of both the parent metabolites (molecular ions in MS 1 ) and fragments (in MS 2 ) revealed differential arsenite effects on individual enzyme reactions. For example, the first two products of glutaminase (i.e., 13 C 5 , 15 N-Glu in a and 13 C 5 -αKG in b) showed arsenite-induced depletion, which suggests glutaminase inhibition by arsenite. However, from the MS 2 data, we saw 13 C 3 (3)-and 13 C 4 (4)-C1 to C5-citrate (i) accumulated while the product 13 C 3 (3)-and 13 C 4 (4)-C1 to C4-αKG depleted (c), which points to additional block at the aconitase and/or isocitrate dehydrogenase steps. The former is consistent with the known inhibition of aconitase by arsenite (24). If this were the only effect of arsenite, we would expect the same trend for the MS 1 data for citrate (h), which was not the case. The production of these fragments had a contribution from the ME ( , light blue box) and/or PC ( , green box) in addition to the glutaminase ( , red box)-mediated pathways. The observed discrepancy between MS 1 and MS 2 data could be attributed to the confounding activation of the ME and PC-mediated pathways by arsenite, leading to the accumulation of the three citrate fragments. This interpretation could also apply to the discrepancy between MS 1 (f) and MS 2 (g) data of malate. The accumulation of 13 C 4 -succinate (d) and depletion of the products 13 C 4 -fumarate (e) are consistent with the inhibition of succinate dehydrogenase (SDH) based on the MS 1 data, which was reported previously (25). Moreover, the arsenite-induced accumulation in the 13 C 5 , 15 N 1 -Glu (n), and  13 C from the first turn of the PDH, PC, and ME-mediated Krebs cycle reactions, respectively. The X-axis refers to the number of 13 C and/or 15 N atoms in each isotopologue. The Y-axis represents μmole or ion intensity normalized to C 2 , 15 N 1 -Gly moieties (o) of GSH argue for the activation of the GSH synthesis pathway while that in 15 N 1 -Glu suggests enhanced GOT activity in addition. The former is consistent with arsenite-induced GSH accumulation and activation of GSH synthesis genes reported for lung epithelial cells (25). Thus, by combining the MS 1 and MS 2 data, it is practical to translate changes in the complex 13 C-and 15 N-labeling patterns of the Krebs cycle metabolites into altered activity of specific enzymes, which would not be reliable based on either MS1 or MS2 data alone.

The PPP and gluconeogenesis
The PPP is a major route for glucose oxidation to produce ribose-5-phosphate (R5P) and NADPH, which are respectively the precursor to nucleotide synthesis and reductant for anabolic and antioxidant metabolism. In this pathway, [ 13 C 6 ]-Glc is converted to ribulose-5-phosphate (Ru5P) via hexokinase, G6P dehydrogenase, and 6-phosphogluconate dehydrogenase, which is then isomerized to R5P (oxidative branch) and epimerized to xylulose-5-phosphate, followed by the transketolase (TKT) and transaldolase (TALDO) reactions to respectively produce sedoheptulose-7-phosphate (S7P) + glyceraldehyde-3-phosphate and fructose-6-phosphate (F6P) and erythrose-4-phosphate (freely reversible nonoxidative branch), respectively ( Fig. 2A). In [ 13 C 6 ]-Glc-traced A549 cells, we saw domination of fully 13 C-labeled isotopologues of G6P (a), 6PG (b), Ru5P/R5P (c), and S7P (d) in the MS 1 data ( Fig. 2A). For S7P, the 13 C 2 -and 13 C 5 -isotopologues were also present and at higher levels than the 13 C 1 -(absent) and 13 C 3 -isotopologues. Based on the TKT and TALDO reaction mechanism (denoted by green arrows), the former two species can be produced directly by the forward TKT reaction and the latter two species by the reverse TALDO reaction. Thus, the observed scrambled 13 C-labeling patterns of S7P is consistent with higher forward or oxidative PPP than reverse or nonoxidative PPP activity. Selenite treatment enhanced the levels of 13 C 2 -and 13 C 5 -S7P while reducing those of 13 C 1 -and 13 C 3 -S7P (d), which suggests a shift from nonoxidative to NADPH-generating oxidative PPP. This is consistent with the lack of depletion of 13 C-6PG (b) and 13 C-R5P+Ru5P (c) despite the large depletion of G6P (a) by selenite. Interestingly, selenite induced depletion of 13 C 5 -and 13 C 6 -F6P (e) but buildup of the 13 C 3 -4,5,6 fragment of F6P (f). Together with the accumulation of 13 C-labeled S7P, the former points to inhibition of TALDO activity by selenite while the latter could be attributed to enhanced gluconeogenesis by selenite (cf., Fig. S3).
In [ 13 C 5 , 15 N 2 ]-Gln-traced BEAS-2B cells, very low levels of 13 C incorporation were evident in some of the PPP products and their 13 C scrambling patterns presumably resulted from a combination of gluconeogenic, TKT, and TALDO activities (Fig. 2B). The fully 13 C-labeled isotopologues of G6P (a), R5P+Ru5P (c), and F6P (e) as well as 13 C 1 -6PG (b) accumulated more in BAsT than control cells. Although most of these changes were at the detection limit and nonstatistically significant, they could reflect enhanced oxidative PPP activity in BAsT cells (cf., Fig. 2A). This would generate more NADPH to support reduction of GSSG to GSH (cf., Fig. 1B) for relieving oxidative stress induced by arsenite (25).

UDP-GlcNAc biosynthesis pathway
UDP-GlcNAc is an activated form of GlcNAc needed for O-and N-linked protein glycosylation, which are important in regulating numerous cellular processes, such as protein targeting to organelles (26) and nutrient sensing (27,28). UDP-GlcNAc has four biochemical moieties (Fig. S4) that are derived from several intersecting metabolic pathways (29) (Fig. 3). The hexosamine moiety comes from glucose and the amido N of Gln via the hexosamine biosynthesis pathway (HBP), the acetyl group is donated from acetyl CoA generated from glucose, amino acids, or fatty acids, the ribose unit derives from glucose via the PPP, and the uracil ring is produced from pyrimidine biosynthesis using C and N sources such as glucose and Gln.
Again, such detailed deduction of selenite's effect on the UDP-GlcNAc biosynthetic pathway would not be feasible without the combined MS 1 and MS 2 data.

Discussion
We have applied a previously developed Ion chromatography-ultrahigh resolution Fourier transform MS 1 / DI-MS 2 method (21) for extensive and robust reconstruction of [ 13 C 6 ]-Glc or [ 13 C 5 , 15 N 2 ]-Gln-fueled central metabolic networks in mammalian cells. This method met the needs for resolving dual tracer distribution in intact metabolites with ultra high-resolution MS 1 while simultaneously acquiring positional labeling in metabolite moieties via DI-MS 2 . In this report, we illustrated how to rigorously reconstruct the Krebs cycle, PPP, gluconeogenesis, and UDP-GlcNAc synthesis pathway by utilizing the combination of UHR-MS 1 with MS 2 data. This approach enabled us to unambiguously discern incell-altered activity of specific enzymes induced by anticancer selenite treatment in lung adenocarcinoma A549 spheroids or by arsenite transformation in lung epithelial BEAS-2B cells.
For A549 spheroids, we found that selenite's ability to attenuate the Krebs cycle activity lies in the blockade of enzymes both in the canonical (OGDH) and anaplerotic (PC, ME) pathways (Fig. 1A). This is consistent with the suppression of the OGDH gene and PC protein but contrary to the overexpression of the ME gene in the 2D counterparts reported previously (13,30). Another notable distinction of selenite's effect is less inhibition of GSH synthesis in 3D (Fig. 1A) versus 2D A549 cells (15), which should contribute to a better capacity of the spheroid culture for antioxidation. Our present data points to reduced synthesis (i.e., blocked GOT), rather than attenuated incorporation, of the precursor Glu as the cause for selenite's inhibition of GSH synthesis in A549 spheroids. This is reasoned from the depletion of 13 C-labeled Glu despite the buildup of its 13 C-labeled αKG precursor. As for PPP, selenite-induced shift to the oxidative branch is expected to produce more NADPH to better sustain the reduction of GSSG to GSH, which is used to alleviate oxidative stress by detoxifying reactive oxygen species (15). This shift can also maintain R5P production despite the block of the TALDO activity in the nonoxidative branch ( Fig. 2A). These changes of the GSH and R5P synthesis pathways in 3D A549 spheroids presumably contribute to their better resistance to selenite toxicity than the 2D counterpart, as observed previously (15). In addition, our combined MS 1 and MS 2 data revealed that subsequent R5P incorporation into UTP and the supply of acetyl CoA and/or its entry into HBP was blocked by selenite, leading to attenuated synthesis of UDP-GlcNAc. This, together with somewhat compromised Krebs cycle, could underlie the growth inhibition of A549 spheroids with prolonged selenite treatment (15).
Arsenite is known to impact various metabolic proteins that contain the sulfhydryl group (31) (e.g., IκB kinase and glucose transporter) leading to different disease states including cancer (32,33). However, the details of metabolic reprogramming in transformed epithelial cells induced by chronic, low-dose exposure to arsenite are still elusive. Our MS 1 -and MS 2based metabolic network reconstruction revealed the complex action of arsenite on the Krebs cycle, PPP, and antioxidation pathways in lung epithelial BEAS-2B cells, including blockade of aconitase, isocitrate dehydrogenase, SDH, and glutaminase but activation of ME/PC, GOT, and GSH synthesis activities. One important outcome of these reprogrammed events can be reactive oxygen species buildup but not in excess to avoid apoptosis while driving different carcinogenic events (33). Moreover, despite the block of HBP and overall uracil synthesis, arsenite-transformed BEAS-2B cells largely maintained UDP-GlcNAc production by activating the CP synthesis and/ or incorporation steps of the UDP-GlcNAc synthesis pathway. UDP-GlcNAc is the required substrate for O-GlcNAcylation of several oncogenic regulators that drive cancer development (34) and the maintenance of this oncometabolite pool is expected to be important to arsenite transformation of BEAS-2B cells.
In conclusion, we applied an IC-UHR-MS 1 /DI-MS 2 method to track changes in 13 C/ 15 N-labeling patterns of metabolites and their moieties in SIRM studies of A549 spheroids or BEAS-2B cells in response to selenite or arsenite transformation, respectively. This approach enabled robust reconstruction of the metabolic network consisting of the Krebs cycle, PPP, gluconeogenesis, and UDP-GlcNAc synthesis pathway to discern specific enzyme activities in the network altered by the treatments. In turn, this information helps elucidate the resistance mechanism of 3D versus 2D A549 cultures to selenite and metabolic reprogramming that can mediate the transformation of BEAS-2B cells by arsenite.

Materials
All materials including the make-up solvent methanol for Ion chromatography, individual standards of metabolites used for quantification were obtained as described previously (21).

Preparation of calibration standard mixtures
A mixture of 86 (Mix 1) and 81 (Mix 2) standards were prepared as two separate calibration standard mixtures as described previously (21). The standard mixtures were aliquoted, lyophilized, and stored at −80 C for long term use. When needed, lyophilized Mix 1 was dissolved in 120 μl 18 MΩ water, vortexed, and 50 μl was used to reconstitute with Mix 2 to form the final calibration standard mixture.

IC-UHR-MS 1 and DI-MS 2
Ion chromatography-ultrahigh resolution fourier transform MS Metabolites were separated on an IonPac AG11-HC-4 μm guard column (2 × 50 mm) coupled to an IonPac AS11-HC-4 μm RFIC&HPIC (2 × 250 mm) analytical column in a Dionex ICS5000 + system (Thermo Scientific) equipped with a dual pump, an eluent generator, an autosampler, and a detector/ chromatography module. Conditions for chromatographic separations (i.e., KOH gradient) and ion suppressor and desolvation in the heated electrospray were as described previously (21). MS data were acquired using the Xcalibur software. A batch of samples started with a 15 min blank (water) injection to check for contamination in the instrument, followed by two injections of calibration standard mixtures to ensure the stability of MS signals and another 15 min water injection to check for carryover on the IC column. Lyophilized cell extracts were freshly reconstituted in 20 μl 18 MΩ water plus 1 μM DSS (sodium trimethylsilylpropanesulfonate) and run in a random order. Each sample was followed by one or two 15 min injections of water blank to minimize carryover. The calibration standard mixture was run after every 6 to 8 cell extracts to track signal loss in the same batch of run. Each sample batch ended with an injection of the calibration standard mixture, followed by water to double check the normality of MS signals and sample carryover.

DI-MS 2 measurement for cell polar extracts
DI-MS 2 analysis was performed in between full MS 1 scans for quantifying targeted fragment(s) of major metabolites in polar extracts, as described previously (21). To achieve this, we set (1) the cycle time of no more than 2 to 3 s for acquiring 10 to 15 points across each chromatographic peak for reliable quantification of precursors and their isotopologues; (2) sufficient resolving power in full scan (500,000) and MS 2 (60,000) modes to discriminate 13 C from 15 N-containing isotopologues of precursors and fragments; and (3) full isotopologue coverage for each metabolite in selecting the precursor mass range for MS 2 scan (i.e., 280-440 with the isolation window of 200 m/z). Other conditions were as described previously (21).

Data analysis and quantification
We first established an in-house exact mass database for the precursors and fragment products based on the corresponding mass ion spectra acquired for individual metabolite standards. Several public metabolomics databases, including the Human Metabolome DataBase (35), the Kyoto Encyclopedia of Genes and Genomes (36), and METLIN (37), and Mass Frontier were used to help interpret MS 2 data for metabolite fragmentation patterns. This database was then incorporated into Trace-Finder v3.3 (Thermo Scientific) for assigning and integrating the peak areas of precursor ions in MS 1 spectra and fragment ions in MS 2 spectra of targeted metabolites in cell extracts for further quantification. Precursors and fragments were assigned with mass accuracy set to 5 ppm. Assignments were curated before isotopic peak areas were corrected for natural abundance as previously described (38). Metabolites in samples were quantified from the corrected MS 1 data by calibrating against the two calibration standard mixtures run before (Std 1) and after (Std 2) the samples. The response factor was calculated for each sandwiched sample run as follows: Response factor = (Area [Std 1] + (Area [Std 2] -Area [Std 1]) × nth run number/run number))/std concentration. The metabolite concentration was then calculated by dividing the corrected MS 1 peak area with the response factor and normalized against the extract aliquot and amount of total protein. The fragment peak areas were similarly normalized.
Preparation of 13 C-labeled polar extracts of 3D A549 spheroids ± selenite A549 cells were grown to 90% confluence in 10-cm plates, followed by loading with magnetic nanoparticles (Nanoshuttle, N3D Biosciences) overnight at 37 C/5% CO 2 , as described previously (15). Cells were then detached and seeded into 6well Costar-cell repellent plates (Corning, Inc) at 400,000 cells/well for spheroid formation. Spheroids were cultured for 4 days before medium change to [ 13 C 6 ]-Glc ± 10 μM Na 2 SeO 3 and grown at 37 C/5% CO 2 for 24 h. Spheroids were harvested, rinsed twice with cold PBS, and then briefly with cold nanopure water before simultaneous quenching and extraction of polar metabolites in cold 70% methanol (15). One-eighth of the polar fraction was aliquoted and lyophilized for IC-UHR-MS 1 /DI-MS 2 analysis.

Data availability
All data acquired are available upon request.